Upload app.py with huggingface_hub

app.py

@@ -14,7 +14,6 @@ from transformers import BertTokenizer, BertForSequenceClassification
 import os
 
 def load_model():
-    # Use BertForSequenceClassification instead of AutoModelForSequenceClassification
     model_name = "indobenchmark/indobert-lite-large-p2"
     model = BertForSequenceClassification.from_pretrained(
         model_name, 
@@ -23,18 +22,16 @@ def load_model():
         ignore_mismatched_sizes=True
     )
 
-    # Try to load local weights if available
     try:
-        # Option 1: Load from local safetensors file
-        local_model_path = "model.safetensors"
+        local_model_path = "model.pt"  # Changed from .safetensors
         if os.path.exists(local_model_path):
             weights = torch.load(local_model_path)
             model.load_state_dict(weights)
+            print("✅ Local model weights loaded successfully!")
         else:
-            print("Warning: No local model weights found. Using pre-trained weights.")
+            print("❌ No local model weights found. Using pre-trained weights.")
     except Exception as e:
-        print(f"Error loading local weights: {e}")
-        print("Falling back to pre-trained model.")
+        print(f"❌ Error loading local weights: {e}")
 
     model.eval()
     return model
@@ -49,13 +46,25 @@ def predict_stress_with_accuracy(text_input):
     if not text_input.strip():
         return None, None, None, None
 
+    # Print input for debugging
+    print(f"🔍 Input Text: '{text_input}'")
+
     # Tokenize the input text
     inputs = tokenizer(text_input, return_tensors="pt", padding=True, truncation=True, max_length=512)
+    
+    # Print tokenization details
+    print("🔍 Tokenization Details:")
+    print(f"  Input IDs: {inputs['input_ids']}")
+    print(f"  Attention Mask: {inputs['attention_mask']}")
 
     # Get the model's output
     with torch.no_grad():
         output = model(**inputs)
 
+    # Print raw logits
+    print("🔍 Raw Logits:")
+    print(output.logits)
+
     # Apply softmax to get probabilities
     probabilities = F.softmax(output.logits, dim=1)
 
@@ -63,6 +72,12 @@ def predict_stress_with_accuracy(text_input):
     probs = probabilities[0].tolist()
     confidence_scores = [round(p * 100, 1) for p in probs]
 
+    # Print probabilities
+    print("🔍 Class Probabilities:")
+    print(f"  Neutral: {confidence_scores[0]}%")
+    print(f"  Mild Stress: {confidence_scores[1]}%")
+    print(f"  Very Stress: {confidence_scores[2]}%")
+
     # Get main prediction
     predicted_class = torch.argmax(probabilities, dim=1).item()
     main_confidence = confidence_scores[predicted_class]
@@ -71,7 +86,9 @@ def predict_stress_with_accuracy(text_input):
     stress_levels = {0: "Neutral", 1: "Mild Stress", 2: "Very Stress"}
     prediction = stress_levels[predicted_class]
 
-    # Generate HTML for the main result
+    print(f"🔍 Predicted Class: {prediction} ({main_confidence}% confident)")
+
+    # Rest of the HTML generation code remains the same...
     result_html = f"""
         <div class="result-card">
             <div class="prediction-text">{prediction}</div>
@@ -82,7 +99,6 @@ def predict_stress_with_accuracy(text_input):
         </div>
     """
 
-    # Generate HTML for detailed analysis
     detailed_html = f"""
         <div class="detailed-analysis">
             <div class="analysis-title">Detailed Analysis</div>